Multiversion

Multiversion is a design principle in data management and software systems that maintains multiple historical versions of an item to support concurrent access, historical queries, and non-blocking reads. The term is most commonly associated with multiversion concurrency control (MVCC) in database systems.

In MVCC, each data item can have several versions, each associated with a version identifier such as a transaction timestamp or a logical clock. When a transaction starts, it views a consistent snapshot corresponding to its version identifier, even as other transactions write new versions. Writes create new versions rather than overwriting existing ones, enabling readers to proceed without waiting for writers in many cases. This approach reduces locking and can improve read throughput and support long-running reads.

To manage storage, systems employ a garbage collection or version-pruning process to remove obsolete versions that are no longer visible to any active transaction. The exact rules for visibility and cleanup depend on the implementation and the chosen consistency guarantees.

Advantages of multiversion approaches include non-blocking reads, finer-grained isolation, and support for time-travel or historical queries. They are widely used in relational databases such as PostgreSQL and Oracle, as well as in various NoSQL systems and distributed data stores that prioritize concurrent access.

Challenges include increased storage overhead, the need for sophisticated garbage collection, potential complexity in conflict resolution, and ensuring correct visibility across distributed components. Proper tuning and workload understanding are important to reap the benefits of multiversion techniques.